Booking.com has shared information about AgentHub, an internal platform that provides modular artificial intelligence (AI) agents for its teams.
AgentHub’s configuration-driven setups and shared infrastructure allow teams at Booking.com, such as user experience writers, product managers and developers, to build and iterate on agents faster, Booking.com's Moran Beladev, Hadas Harush and Chana Ross wrote in a blog post outlining the platform.
According to the blog post, with AgentHub, teams can deploy AI agents “in a matter of hours.”
“As we started to build and ship more and more GenAI based use cases, we had to ask a more fundamental question: Should every team really solve all of this again, for each new use case?” Guy Nadav, director of machine learning and AI at Booking.com, wrote on LinkedIn.
“This led us to develop AgentHub, our platform for building modular, production-grade AI agents and workflows at Booking.com. The idea is relatively simple—let the agent reason about a task and use tools that expose existing capabilities: retrieval, property data, recommendations, content intelligence, internal APIs or external services.”
In a series of LinkedIn posts, Nadav has described Booking.com’s work with machine learning and AI platforms for ranking, recommendations and content intelligence, clarifying that AgentHub builds on these platforms.
“We simply added a new layer that can reason about when and how to use them,” Nadav wrote in his most recent post.
The blog post states that AgentHub has use cases across the booking funnel, from discovery and planning to booking to cross-selling and trip management.
One of the first agents built on AgentHub, the AI Trip Planner Q&A, answers travelers’ property-related questions.
The Q&A agent matches property mentions from users with the correct Booking.com property ID, fetches relevant information from internal systems and calculates distances and travel times to places such as landmarks and transportation hubs. Booking.com rebuilt an existing deterministic flow, now letting the agent choose which systems to call.
As a result, Booking.com said it saw improvements in answer and context relevance and factual accuracy, as well as a decrease in large language model rejection rate.
The blog authors added the platform is still evolving, and Booking.com plans to offer multi-agent collaboration, multimodal capabilities for images and other non-text inputs, long-term memory and personalization, enhanced tool-calling methods and agent-harness optimization.
Expedia Group chief AI and data officer Xavier Amatriain also recently outlined the company’s operating model for AI, defining an agent as “an AI-powered system that works toward a goal.”
In his blog post, he also spoke of the future of travel AI as a “composed, multi-agent system.”
He said specialized agents could deliver value by carrying out specific jobs such as helping with discovery, comparing properties, building itineraries, answering questions or helping travelers with trip changes.